CNN-based search model underestimates attention guidance by simple visual features

TL;DR

This paper evaluates a CNN-based attention guidance model in visual search tasks and finds it underestimates human attention guidance due to lack of bottom-up guidance and potentially inadequate feature learning.

Contribution

The study adapts a CNN-based attention model for search experiments and demonstrates its limitations in replicating human attention guidance.

Findings

CNN model underestimates human attention guidance

Lacks bottom-up guidance in the model

Standard CNNs may not learn features needed for human-like attention

Abstract

Recently, Zhang et al. (2018) proposed an interesting model of attention guidance that uses visual features learnt by convolutional neural networks for object recognition. I adapted this model for search experiments with accuracy as the measure of performance. Simulation of our previously published feature and conjunction search experiments revealed that CNN-based search model considerably underestimates human attention guidance by simple visual features. A simple explanation is that the model has no bottom-up guidance of attention. Another view might be that standard CNNs do not learn features required for human-like attention guidance.